Sage Journals: Discover world-class research

Abstract

Objective: To examine the applications of artificial intelligence (AI) in lifestyle medicine focused on diabetes care as a narrative review. Methods: Relevant keywords were identified and searched using PubMed to find relevant studies on AI in diabetes lifestyle management. Results: AI applications in diabetes care were divided into four primary categories: 1- predictive models for diabetes risk and complications, which can utilize random forest and deep learning, demonstrating high accuracy rates (>80%); 2- personalized lifestyle recommendations, which can utilize clustering techniques and causal forest analysis to adapt interventions, leading to enhanced glycemic control and weight reduction; 3- remote monitoring and self-management tools, which can utilize digital twin technology and machine learning for behavior modeling, showing improved patient adherence and clinical results; and 4- clinical decision support systems, which assess various data sources for enhanced diagnosis and treatment suggestions. Conclusion: AI technologies show significant potential in improving diabetes care through multiple modalities that offer scalable cost-effective solutions, improved patient outcomes, and more efficient resource distribution.

Keywords

artificial intelligence clinical decision support deep learning diabetes care lifestyle medicine machine learning personalized interventions

“The DBCD model stresses the importance of early intervention at all chronic disease stages through structured lifestyle changes, with judicious pharmacotherapy/procedures at later stages.”

Introduction

The escalating prevalence of diabetes poses a substantial burden on individuals, healthcare systems, and global economies. The number of adults living with diabetes worldwide has surpassed 800 million in 2022, more than quadrupling since 1990.¹ This significant rise is largely attributed to factors such as increased obesity rates, unhealthy diets, lack of physical activity, and economic hardship.¹ Pharmacological interventions are crucial, but lifestyle modifications are paramount in mitigating chronic disease progression and associated complications.² However, implementing and maintaining effective lifestyle changes can be challenging for many individuals, theoretically driven by complex interactions among primary drivers of chronic disease (genetics, behavior, and environment), and pragmatically driven by the networking effects of technological, social, and other infrastructural factors.^3-5

Artificial Intelligence (AI) offers a promising approach to optimizing lifestyle medicine in patients with diabetes. AI encompasses various techniques, including machine learning (ML), deep learning (DL), and natural language processing (NLP).⁶ In the context of diabetes care, these AI algorithms analyze extensive datasets—such as patient data, glucose readings, and lifestyle information—to achieve several key functions: predicting individual risk, tailoring lifestyle recommendations (e.g., for nutrition and physical activity), facilitating remote monitoring, automating support tasks (e.g., alerts, prompts), and providing clinical decision support.⁷ Machine learning, in particular, is foundational to many such systems, enabling algorithms to learn patterns from data to make these predictions and personalize interventions. Table 1 presents a glossary of AI tools recruited for this narrative review.

Table 1.

Glossary of Artificial Intelligence Tools.^a

Method	Definition
Artificial Neural Networks (ANN)	A computational model consisting of layers of interconnected nodes inspired by the way biological neural networks in the human brain process information. ANNs are used for various tasks, including classification and regression
Bayesian Belief Networks (BBN)	A probabilistic graphical model that represents a set of variables and their conditional dependencies via a directed acyclic graph. BBNs are used for reasoning under uncertainty
Convolutional Neural Network (CNN)	A type of deep learning model particularly effective in processing structured grid data, such as images. CNNs automatically detect features and patterns in visual data
Decision Tree	A decision support tool that uses a tree-like model of decisions and their possible consequences. It is commonly used for classification and regression tasks
Deep Learning (DL)	A subset of machine learning that uses neural networks with multiple hidden layers (deep networks) to analyze various forms of data, including images, audio, and text. It enables complex operations on massive amount of data
Digital Twin (DT)	A digital replica of a physical entity or system that can be used for simulation, analysis, and monitoring in real-time; in healthcare, DTs leverage nudges/prompts, as well as incorporating human touches
Ensemble Learning Methods	Ensemble learning is a powerful machine learning technique that enhances predictive performance by combining multiple models
Expectation-Maximization (EM)	Iterative algorithm used to find maximum likelihood estimates of parameters in statistical models with incomplete data. It alternates between an expectation (E) step, which estimates missing data, and a maximization (M) step, which updates the parameters, continuing until convergence
Extreme Gradient Boosting (XGBoost)	Represents a decision tree-based ensemble algorithm that uses a gradient boosting framework designed to optimize performance and speed by creating multiple weaker models and combining them
Gradient Boosting Decision Tree (GBDT)	A machine learning technique that builds models in a stage-wise fashion by combining weak learners to improve predictive accuracy
K-Nearest Neighbor (KNN)	A simple, non-parametric algorithm used for classification and regression that predicts the output based on the k closest training examples in the feature space
Light Gradient Boosting Machine (LightGBM)	A gradient boosting framework that uses tree-based learning algorithms, optimized for speed and efficiency, especially with large datasets
Long Short-Term Memory (LSTM)	A type of recurrent neural network architecture designed to learn long-term dependencies in sequential data, making it effective for tasks like time series prediction and natural language processing
Logistic Regression (LR)	A statistical method used for binary classification that models the probability of a binary outcome based on one or more predictor variables
Machine Learning (ML)	A branch of artificial intelligence that focuses on building systems capable of learning from data and improving their performance over time without being explicitly programmed
Multilayer Perceptron (MLP)	A type of feedforward artificial neural network consisting of multiple layers of nodes, where each layer is fully connected to the next one
Natural Language Processing (NLP)	A field at the intersection of computer science and linguistics that enables machines to understand, interpret, and generate human language
Nudge	A subtle instruction (human or technological) to change user behavior with a predicted effect but also with the element of user choice
Prompt	A direct instruction (human or technological) to perform a specific task with a predicted specific result but also with the element of user choice
Random Forest (RF)	An ensemble learning method that constructs multiple decision trees during training time and outputs the mode or mean prediction of individual trees for improved accuracy
Reinforcement Learning	An area of machine learning where an agent learns to make decisions by taking actions in an environment to maximize a cumulative reward over time
Shapley Additive Explanations (SHAP)	A method to explain individual predictions by attributing the contribution of each feature to the overall prediction using cooperative game theory principles
Support Vector Machines (SVM)	Supervised learning models used for classification tasks by finding the hyperplane that best separates different classes in the feature space
Support Vector Regression (SVR)	An extension of SVM for regression tasks that aims to find a function that deviates from actual target values by a value no greater than a specified margin
Tree-based Pipeline Optimization Tool (TPOT)	An automated machine learning tool that optimizes machine learning pipelines using genetic programming techniques

^aThis glossary is limited to terms used in this narrative review and should not be interpreted as a complete listing.

The application of AI in diabetes care is particularly compelling due to the ability to address the multifaceted and dynamic aspects of chronic disease. For instance, AI can enhance diabetes care through continuous glucose monitoring (CGM) interpretation and insulin dosing (using insights from multivariate changes over time, such as variations in insulin sensitivity at different times of the day with different meals and varying glycemic indices/loads).⁸ Comprehensive AI systems can integrate data on dietary instructions, physical activity, stress levels, and comorbidities to personalize diabetes management strategies. This approach is particularly relevant for type 2 diabetes (T2D) but can also apply to type 1 diabetes (T1D) and gestational diabetes mellitus (GDM).⁹ However, despite the burgeoning research, a dedicated review that holistically synthesizes the role and evidence base of diverse AI technologies specifically in empowering lifestyle medicine for comprehensive diabetes management is currently lacking. This narrative review aims to identify and critically evaluate the principal applications of AI in supporting key aspects of lifestyle medicine for diabetes care. Specifically, it examines how AI is utilized in (1) predictive modeling for risk assessment, (2) personalized lifestyle guidance (nutrition and physical activity), (3) remote monitoring and patient self-management, and (4) clinical decision support relevant to lifestyle interventions. The review synthesizes existing scientific evidence on the characteristic features of these AI technologies and their reported effectiveness in improving diabetes-related health outcomes and enhancing patient self-management capabilities. We have organized this review by AI-enabled technologies/methods and their respective scientific evidence.

Methods

This study employs a narrative review methodology to provide a critically evaluative overview of the multifaceted applications of Artificial Intelligence (AI) in lifestyle medicine for diabetes care. This approach was chosen to facilitate an in-depth synthesis of key technological features, evidentiary findings, and overarching trends.

To identify relevant literature, a primary search was conducted in the PubMed database, selected for its comprehensive coverage of peer-reviewed biomedical and clinical research directly pertinent to healthcare applications. We acknowledge that this focus, while prioritizing studies with a certain level of scientific scrutiny, may not capture all engineering-specific publications or gray literature potentially found in specialized AI/computer science databases. The search strategy in PubMed combined MeSH (Medical Subject Headings) terms and free-text keywords using Boolean operators (AND, OR). Core search concepts encompassed AI-related terms (e.g., “Artificial Intelligence,” “Machine Learning,” “Deep Learning,” “Natural Language Processing,” “predictive model*,” “Algorithms,” “Clinical Decision Support”), diabetes-related terms (e.g., “Diabetes Mellitus,” “Type 1 Diabetes,” “Type 2 Diabetes,” “Gestational Diabetes,” “Prediabetes”), and lifestyle medicine-related terms (e.g., “Lifestyle,” “diet,” “nutrition,” “physical activity,” “behavior change,” “Self-Management,” “remote monitoring,” “personalized intervention*”). An exemplary PubMed search combined these categories, such as (AI terms) AND (Diabetes terms) AND (Lifestyle terms). Furthermore, the reference lists of retrieved articles, particularly influential reviews and key primary research studies, were manually scanned to identify additional relevant publications (a process often referred to as snowballing or citation searching) to complement the database search.

Studies were selected for inclusion if they directly investigated, applied, or discussed the use of AI technologies in supporting lifestyle medicine components (including risk prediction informing lifestyle modification, dietary management, physical activity promotion, behavioral support, remote monitoring for self-management, and clinical decision support systems focused on lifestyle aspects) for individuals with any type of diabetes (type 1, type 2, gestational) or prediabetes, and involved human participants. Articles had to be published in the English language. While no strict limitations were placed on study design to capture a broad spectrum of evidence, the primary search focus was on publications from January 2017 to May 2024 to reflect recent advancements, though seminal earlier works providing critical foundational context were also considered.

The initial search results were screened by title and abstract. Full texts of potentially eligible articles were then retrieved and assessed against the inclusion criteria. Given the narrative nature, the selection aimed to identify key themes and representative examples of AI applications. Data on AI technology, characteristic features, target population, key findings, and study limitations were qualitatively extracted and synthesized. The results are presented thematically, organized by the main AI-enabled application areas identified, and further categorized by their characteristic features and key evidentiary findings, as outlined in the review’s aim.

Results

Predictive AI Models

Characteristic Features

Artificial intelligence algorithms can predict an individual’s risk of developing type 2 diabetes (T2D) and potential complications. Crucially, for lifestyle medicine, these predictions serve as an essential early warning, enabling the timely initiation and tailoring of preventative lifestyle strategies and interventions before clinical disease manifestation or progression (Table 2).¹⁰ These predictive models use ML techniques to analyze various data points, including genetic and clinical data.¹¹

Table 2.

Studies Published on Predictive Models and Risk Assessment.^a

Reference (Year, Country)	Aim	Design and Population	AI Tool and Methods	Results	Conclusions
González-Martín et al²⁰ (2023, Spain)	To model insulin sensitivity, resistance, and diabetes using microarray data and ML	Study from Gene Expression Omnibus; patients with diabetes, sensitivity, and resistance	Bioconductor in R; PCA; ML techniques (MLP, kNN, ANN, SVM, RF) with 1000 iterations	Found 20 differentially expressed genes to classify patients; models had accuracy >80%	ML techniques effectively classify insulin-related conditions with high accuracy
Wandell et al²¹ (2024, Sweden)	To predict diabetes using ML in primary health care	Case-control; 311,232 individuals with diabetes and controls	SGB technique; Bernoulli loss function; prediction with odds ratios of marginal effects (ORME)	Hypertension and obesity were major predictors	Effective screening could involve known risk factors like hypertension, obesity; ML improves prediction
Okada et al⁴⁷ (2022, Japan)	To predict follow-up visit failure using ML	Retrospective cohort; 10,645 patients	Lasso regression with Lasso regularization	Four predictors identified; model had discrimination ability of 0.71 C-statistic	Effective ML-based models help identify patients likely to miss follow-ups
Maqbool et al¹⁷ (2023, Pakistan)	To monitor diabetes patients using AI and IoT architecture	Validated system on 50 diabetes patients	Sensor data, fuzzy logic, ML algorithms (e.g., RF)	High prediction similarity with standard methods	The system predicts health conditions of patients with diabetes and provides noninvasive monitoring
Borzouei et al⁴⁸ (2018, Iran)	To identify key demographic risk factors for T2D using a neural network model	Cross-sectional; 234 individuals diagnosed with T2D	ANN with 3 layers and Broyden-Fletcher-Golfarb Shanno algorithm	Waist circumference and age among key predictors	Identifies demographic risk factors efficiently for high-risk T2D groups
Elhadd et al⁴⁹ (2020, Qatar)	To predict glucose levels during Ramadan using ML models	Cohort with 13 patients and monitoring devices	Random Forest, XGBoost, and deep learning techniques	Accurately predicted normal and hyperglycemic excursions; limited for hypoglycemia	XGBoost performs well for predicting normal and hyperglycemic events; less effective for hypoglycemia prediction during fasting
Naveed et al¹⁴ (2023, Canada)	To assess the usefulness of deep learning models in predicting diabetes considering time-dependent risks	Longitudinal; over 19,000 patients’ EMR from the Canadian PCSSN	LSTM, CNN, and hybrid models; Adam optimization	Achieved >91% accuracy; significant risk factors identified	Confirms LSTM handles time-dependent risk well, improving diabetes prediction performance
Kim YH et al¹⁹ (2024, South Korea)	To use AI-based analysis of CT images for diabetes risk prediction	Cross-sectional and longitudinal; 15,330 subjects with CT images	AI-DeepCatch software to analyze body composition metrics	Higher VF proportion and VF/SF ratio predictive of diabetes	AI-based CT image analysis effectively predicts diabetes risk through body composition metrics
Kumar et al¹⁸ (2022, Singapore)	To create a GDM predictor for early intervention	Prospective cohort; 222 women in S-PRESTO study	AutoML using TPOT with preconception features; cross-validation	High AUC of 0.93; identified preconception predictors	Early GDM detection model facilitates targeted intervention; as part of digital health strategies
Choi SG et al¹⁵ (2023, South Korea)	To compare ML-based diabetes prediction models to traditional models	Cross-sectional with 32,827 from KNHANES 2014-2020 data	Ensemble learning techniques with feature evaluation (SHAP framework)	ML models had consistently higher AUC than traditional models	ML-based models enhance prediction of undiagnosed diabetes over traditional statistical methods
Das et al¹² (2024, Bangladesh)	To understand T2D’s influence on CVD risk and develop precise prediction models	Cohort using Bangladesh Demographic and Health Surveys	Various ML models (SVM, DT, RF, NB, KNN, LightGBM, XGBoost)	Highlighted key age and weight-related CVD risks; RF showed excellent specificity	ML models contributed effectively to CVD risk assessment and intervention strategies targeting risk factors
Neri-Rosario et al⁵⁰ (2023, Mexico)	To use ML to classify taxa in individuals with prediabetes or T2D	Cohort of 410 patients categorized as normoglycemic, prediabetic, or T2D	ML models (Logistic Regression, NB, DT, RF, XGBoost, MLP) compared	Random Forest showed highest predictive performance	Highlights role of gut microbiota and its modulation in predicting and managing diabetes stages
Kim R et al⁵¹ (2023, South Korea)	To assess diabetes and comorbid conditions in middle-aged/older adults	Cohort with 5527 participants	Multiple AI approaches (RF, SVM, ANN, etc.) with validation via variable importance analysis	Diabetes/comorbidity strong association confirmed; determinants similar across pairs	Reinforces need for comprehensive comorbidity management in diabetes care
Gallardo-Rincón et al⁵² (2023, Mexico)	To develop AI-based model for GDM risk prediction in Mexican women	Prospective cohort with 1709 pregnant women	ANN with selected clinical/demographic predictors	Model demonstrated 70.3% accuracy and 83.3% sensitivity	Model aids decision-making in healthcare concerning GDM risk without needing lab tests
Watanabe et al⁵³ (2023, Japan)	To evaluate AI-based GDM prediction models using cohort data	Cohort with 82,698 pregnant mothers	Machine learning methods (SVM, RF, GBDT) with LR as a reference	GBDT displayed highest accuracy and interpretability	AI models, particularly GBDT, are highly effective for predicting GDM using diverse metadata
Ou et al¹⁶ (2023, Taiwan)	To develop predictive models for ESRD risk in T2D patients	Cohort with 53,477 newly diagnosed T2D patients	Machine learning with models like XGBoost; addressed imbalance with SMOTE-Tomek	XGBoost achieved highest AUC of 0.953 in identifying ESRD risk	Developed model predicts risk of ESRD effectively, facilitating early intervention
Yao et al⁵⁴ (2023, USA)	To use image-derived data to automate T2D risk prediction without blood tests	Retrospective cohort of 389 patients	Neuroal networks and DT models; evaluated with novel metric for generalization	High sensitivity and effectiveness in predicting diabetes without laboratory tests	Leverages AI for opportunistic risk stratification, offering noninvasive prediction of T2D risk
Hsieh et al¹³ (2019, Taiwan)	To assess models predicting breast cancer risk in T2D patients	Cohort with 636,111 female T2D patients	ANN, LR, and RF models evaluated with cross-validation techniques	RF had largest AUC (0.959); effective prediction with patient-centric factors	RF is highly effective at predicting risk with precision and sensitivity in T2D populations
Haneef et al⁵⁵ (2021, France)	To develop a generic ML algorithm to estimate diabetes incidence using reimbursements	Cohort with 44,659 participants from CONSTANCES	Supervised ML approach (Linear Discriminant Analysis) evaluated by AUC performance measure	Moderate performance with algorithm having 62% sensitivity and 67% accuracy	ML offers innovative opportunities for public health surveillance, though improvements are needed for wider applications
Hennebelle et al⁵⁶ (2023, UAE)	To propose IoT-AI-blockchain system for diabetes prediction based on risk factors	Three datasets from different geographical regions	IoT with RF, LR, and SVM classifiers; comparative analysis of sensors/devices and data processing	RF model yielded highest accuracy among tested models; improves with feature selection	Proposes novel system integrating IoT, AI, and blockchain for reliable, secure diabetes prediction

^aAbbreviations: AI - Artificial Intelligence; ANN - Artificial Neural Network; AUC - Area Under the Curve; CNN - Convolutional Neural Network; CT - Computed Tomography; CVD - Cardiovascular Disease; DT - Decision Tree; EMR – Electronic Medical Record; ESRD - End-Stage Renal Disease; GBDT - Gradient Boosting Decision Tree; GDM - Gestational Diabetes Mellitus; IoT - Internet of Things; KNN - K-Nearest Neighbor; KNHANES - Korea National Health and Nutrition Examination Survey; LSTM - Long Short-Term Memory; LightGBM - Light Gradient Boosting Machine; LR - Logistic Regression; MLP - Multilayer Perceptron; ML - Machine Learning; NB - Naïve Bayes; ORME - Odds Ratios of Marginal Effects; PCSSN - Primary Care Sentinel Surveillance Network; PRESTO - Preconception Study of Long-Term Maternal and Child Outcomes; RF - Random Forest; SHAP - Shapley Additive explanations; SF - subcutaneous fat; SMOTE - Synthetic Minority Over-sampling Technique; SVM - Support Vector Machine; T2D - Type 2 Diabetes; TPOT - Tree-based Pipeline Optimization Tool; VF - Visceral Fat; XGBoost - Extreme Gradient Boosting.

Evidentiary studies spanned diverse geographical regions (Bangladesh, Canada, France, Iran, Japan, Mexico, Pakistan, Qatar, Singapore, South Korea, Spain, Sweden, Taiwan, United Arab Emirates, and the U.S.) from 2018 to 2024. These studies focused on developing predictive models for T2D, gestational diabetes mellitus (GDM), and associated complications such as cardiovascular disease (CVD) and end-stage renal disease (ESRD). The studies employed a variety of AI and ML techniques, including random forest (RF), support vector machines (SVM), artificial neural networks (ANN), long short-term memory (LSTM), and ensemble learning methods. The data sources included electronic medical records (EMR), gene expression data, demographic surveys, and sensor data from Internet of Things (IoT) devices.

There are four main categories of AI/ML techniques used in predictive models in diabetes care. Random forest is widely used for high accuracy and interpretability.^12,13 Deep learning techniques such as bidirectional LSTM and convolutional neural network (CNN) were employed for time-dependent risk prediction.¹⁴ Ensemble learning methods such as XGBoost and LightGBM were effective in improving prediction accuracy.^15,16 Lastly, IoT Integration studies^13,14 combined IoT with AI for noninvasive monitoring and prediction.¹⁷

Key Findings

Many studies reported high accuracy (>80%) in predicting T2D and related conditions. For example, Naveed et al¹⁴ achieved >91% accuracy using LSTM models, and Kumar et al¹⁸ reported an area under the curve (AUC) of 0.93 for GDM prediction, demonstrated by a receiver-operating characteristic (ROC) curve indicating high sensitivity and specificity of the LSTM model for GDM prediction. The significance of such high predictive accuracy lies in its potential to identify individuals who would most benefit from intensive, personalized lifestyle modification programs. For instance, some studies leveraged noninvasive methods, such as computed tomography (CT) image analysis¹⁹ and IoT-based monitoring,¹⁷ to predict diabetes risk without blood tests.

The successful application of AI tools in predictive modeling and risk assessment holds significant implications for proactive lifestyle medicine. Predictive models enable early identification of high-risk individuals, facilitating timely interventions.^18,20 This approach moves beyond reactive care, offering a pathway to integrate AI-driven risk stratification into primary care to promote preventative lifestyle changes, as suggested by Wandell et al^21, who highlighted the role of known lifestyle-related risk factors (hypertension, obesity) in ML-enhanced screening. Ultimately, the “role” of predictive AI in this context is to make lifestyle medicine more targeted, timely, and impactful.

Personalized Lifestyle Guidance

Characteristic Features

AI technologies are increasingly being utilized to provide personalized lifestyle guidance for individuals with diabetes, offering tailored recommendations in nutrition management, physical activity, and overall diabetes care^11,22 (Table 3). These AI-powered systems analyze diverse data sources, such as wearable devices, continuous glucose monitoring (CGM) data, and dietary habits, to deliver actionable insights that improve glycemic control and overall health outcomes.

Table 3.

Studies Published on Personalized Lifestyle Guidance.^a

Reference (Year, Country)	Aim	Design & Population	Methods & AI Tools	Results	Conclusions
Kim KJ et al²³ (2022, South Korea)	To define healthy lifestyle parameters via wearable trackers in T2D patients	Prospective cohort with 24 patients with T2D	EM algorithm for clustering; SHAP for lifestyle factor analysis	Two groups identified; group A had healthier lifestyle indicators and lower HbA1c at follow-up	Regular sleep patterns identified as key lifestyle determinants for better glycemic control using wearable trackers
Baum A et al²⁴ (2017, USA)	To explore heterogeneity in treatment effects of weight loss interventions	RCT with 4901 T2D patients	Causal forest analysis; Cox proportional hazards models	Identified subgroups with distinct HTEs; HbA1c and health surveys help identify these groups for weight loss intervention efficacy	T2D patients with moderate/poor diabetes control benefit from weight loss interventions, while certain others see adverse effects
Kannenberg S et al²⁵ (2024, Germany)	To present a digital therapeutic (DTx) for personalized T2D management	Cohort of 118 participants with non-insulin-treated T2D	Random forest regressor, deep learning for glucose response prediction; CGM data monitored	Improved HbA1c and weight loss across cohorts showing clinical relevance of the DTx	DTx showing effective glycemic control and weight reduction, provides non-pharmacological digital T2D management
Bul K et al²⁶ (2023, UK)	To assess a web-based AI nutrition platform’s usability and efficacy	Cohort with 73 adults with prediabetes or diabetes and carers	AI for personalized meal planning; deep learning and NLP for data processing on diet and shopping habits	Reported weight loss and reduction in waist size despite low regular engagement with the platform	Platform led to perceived good usability; adding educational content could enhance diabetes management comprehension
Graham SA et al²⁷ (2022, USA)	To compare weight loss maintenance of AI-powered DPP users by CDC criteria	Cohort comparing CDC qualifiers (191) and non-qualifiers (223)	AI interface delivering CDC curriculum; logistic regression for weight loss predictors	CDC qualifiers had better weight maintenance; frequent AI-coaching interactions improved weight loss likelihood	AI-enhanced DPP effectively facilitates weight loss, complementing traditional methods with scalable AI solutions

^aAbbreviations: AI - Artificial Intelligence; CDC - Centers for Disease Control and Prevention; CGM - Continuous Glucose Monitoring; DPP - Diabetes Prevention Program; DTx - Digital Therapeutics; EM - Expectation-Maximization; HbA1C - Hemoglobin A1C; NLP - Natural Language Processing; RCT - Randomized Controlled Trial; SHAP - Shapley Additive explanations; T2D - Type 2 Diabetes.

Evidentiary studies spanned diverse geographical regions, including Germany, South Korea, the U.S., and the United Kingdom (U.K.) from 2017 to 2024. These studies focused on leveraging AI to provide personalized lifestyle interventions for individuals with T2D or prediabetes. The research designs included prospective cohorts, randomized controlled trials (RCTs), and digital therapeutics (DTx) evaluations, with sample sizes ranging from 24 to 4901 participants. The AI tools and methods employed in these studies include clustering algorithms, causal forest analysis, RF regressors, DL, and NLP. Data sources encompass wearable trackers, CGM data, dietary logs, and health surveys.

There are five main categories of AI/ML techniques used for lifestyle guidance. Clustering algorithms²³ use the expectation-maximization (EM) algorithm to cluster patients based on lifestyle factors, identifying key determinants for glycemic control. Causal forest analysis²⁴ explores heterogeneity in treatment effects of weight loss interventions, identifying subgroups that benefit most from specific interventions. Kannenberg et al²⁵ utilized RF regressors and DL models to predict glucose responses and evaluate the efficacy of a DTx for T2D management. In addition, Bul et al²⁶ applied NLP and DL to process dietary and shopping habit data, enabling personalized meal planning for patients with diabetes, including T1D and T2D. Lastly, AI-powered coaching²⁷ implements an AI interface to deliver the Centers for Disease Control and Prevention (CDC’s) Diabetes Prevention Program curriculum using logistic regression to identify predictors of weight loss maintenance.

Key Findings

Using AI-enabled wearable devices, Kim et al²³ found that regular sleep patterns were associated with improved glycemic control, while Baum et al²⁴ found that weight loss interventions were associated with improved glycemic control in those with moderate or poor control. Kannenberg et al²⁵ demonstrated improved HbA1c and weight loss through AI and CGM data. Bul et al²⁶ reported weight loss via an AI nutrition platform, even with low engagement, and Graham et al²⁷ showed that AI-powered coaching enhanced weight loss maintenance, particularly for CDC-qualified participants. These findings underscore AI’s potential for tailored, scalable lifestyle interventions in diabetes management.

The integration of AI into personalized lifestyle guidance has significant implications for diabetes care and prevention. For instance, AI-enabled technologies improve the delivery of personalized recommendations, adherence, and outcomes based on individual data.^23,26 Also, DTx and AI-powered platforms offer scalable, cost-effective alternatives to traditional interventions, particularly in resource-limited settings.^25,27 Wearable devices and AI-enabled coaching enhance patient engagement by providing real-time feedback and actionable insights.^23,27 Identifying subgroups that benefit most from specific interventions allows healthcare systems to allocate resources more effectively.²⁴

Remote Monitoring and Self-Management

Characteristic Features

Remote monitoring and self-management technologies leverage AI to empower patients with diabetes to manage their conditions more effectively from their homes (Table 4). These AI-powered systems analyze diverse data sources, such as CGM data, behavioral patterns, and patient inquiries, to provide real-time feedback and support.

Table 4.

Studies Published on Remote Monitoring and Self-Management.^a

Reference (Year, Country)	Aim	Design & Population	AI Tools and Methods	Results	Conclusions
Ansari RM et al²² (2023, Australia)	To evaluate self-management practices in T2D patients in rural Pakistan using AI	Cross-sectional study with 200 T2D patients in rural Pakistan	ANN and LR; dataset split into training (80%) and testing (20%)	Training accuracy 98%; testing accuracy 97.5%; highlighted poor self-management efforts	AI models effectively evaluated self-management activities, indicating a heavy reliance on medication adherence rather than behavioral adjustments
Shamanna P et al²⁸ (2024, India)	To assess DT vs SC effects on BP and HTN in T2D patients	RCT with 319 T2D participants, randomized into DT (233) and SC (86) groups	AI-driven DT and analytics; dietary analysis; BP monitoring	DT group had greater BP reductions and higher normotension and HTN remission rates than SC group	DT technology outperformed SC in reducing BP and inducing HTN remission in T2D patients
Helal A et al³⁰ (2009, USA)	To describe a smart home platform for behavioral monitoring in diabetes management	Integration of various devices for remote monitoring in smart environments	Machine learning for behavior modeling and activity recognition using hidden Markov models	Achieved 98% accuracy in activity recognition	Smart home platform effectively monitors and supports diabetes patients in managing activity, diet, and exercise compliance
Hernandez CA et al³¹ (2023, USA)	To evaluate ChatGPT’s responses to T2D inquiries and assess their reliability	Cross-sectional study using T2D inquiry questions	ChatGPT used to generate responses reviewed by physicians for appropriateness	98.5% of responses were appropriate; highlighted AI reliability over traditional search engines	ChatGPT provides high-quality information for T2D education, with potential for use in patient education enhancements
Williams KJ et al³² (2023, USA)	To report TIR and CGM parameters in long-term d-Nav® users with T2D	Cohort study with 18 T2D patients using d-Nav® app for insulin management	d-Nav® AI adjusts insulin doses based on CGM data	Cohort showed 69.8% TIR, low hypoglycemia, and good glycemic variability	d-Nav® helps maintain recommended glycemic profiles, comparable to automated insulin delivery systems but less intrusive
Nayak A et al³⁴ (2023, USA)	To assess a voice-based AI app for basal insulin titration in T2D patients	RCT with 32 adults needing basal insulin adjustment	Voice-based AI application using Alexa for insulin management	AI group achieved quicker insulin dosing, better adherence, and improved glycemic control	Voice-based AI solutions effectively improve insulin management, glycemic control, and patient adherence at home

^aAbbreviation: AI - Artificial Intelligence; ANN - Artificial Neural Network; BP - Blood Pressure; CGM - Continuous Glucose Monitoring; DT - Digital Twin; HTN - Hypertension; LR - Logistic Regression; RCT - Randomized Controlled Trial; SC - Standard of Care; T2D - Type 2 Diabetes; TIR - Time in Range.

Evidentiary studies spanned various geographical regions, including Australia, India, and the U.S from 2009 to 2024. These studies focused on using AI to improve self-management practices, blood pressure control, and patient education for T2D. The research designs included cross-sectional studies, RCTs, and cohort studies, with sample sizes ranging from 18 to 319 participants. The AI tools and methods employed in these studies included ANN, digital twin (DT) technology, ML for behavior modeling, NLP, and voice-based AI applications. Data sources encompassed CGM data, blood pressure monitoring, smart home devices, and patient inquiries.

There are six main categories of AI/ML techniques used for Remote Monitoring and Self-Management. Ansari et al²² used ANN and logistic regression methods to evaluate self-management practices in rural Pakistan, achieving high accuracy in identifying poor self-management efforts. Shamanna et al²⁸ employed AI-driven Digital Twin (DT) technology and analytics to optimize blood pressure control and increase hypertension remission rates in patients with T2D, demonstrating its superiority over standard care. Additionally, Joshi et al²⁹ demonstrated superior efficacy of DT-enabled personalized nutrition compared with standard of care to improve markers of metabolic dysfunction-associated fatty liver disease. Behavior modeling can be achieved with ML and hidden Markov models to develop a smart home platform (98% accuracy in activity recognition).³⁰ Conversational AI and education tools, such as ChatGPT’s responses to T2D inquiries, highlighted the reliability and potential for patient education.³¹ AI-driven insulin management was part of the d-Nav® app, which adjusts insulin doses based on CGM data, showing equivalency to automated delivery systems in maintaining glycemic control in patients with T2D,³² similar to what was performed by Nimri et al,³³ in T1D management. Lastly, voice-based AI applications have been implemented using Alexa® for basal insulin titration in T2D.³⁴

Key Findings

AI-driven remote monitoring tools and self-management applications are highly effective in supporting diabetes management. For instance, DT technology²⁵ led to greater blood pressure reductions compared to standard care, indicating the value of advanced personalized analytics.²⁸ Other AI-based applications, such as d-Nav® and the voice-based assistant developed by Nayak et al³⁴ demonstrated significant improvements in insulin management and glycemic control, offering scalable solutions for home-based care. Furthermore, the Smart Home Platform reported by Helal et al³⁰ illustrated the practicality of continuous monitoring, and Hernandez et al³¹ highlighted the educational potential of AI in healthcare communication.

The integration of AI into remote monitoring and self-management holds transformative potential for diabetes care. By delivering real-time feedback and personalized support, AI tools enhance patient adherence and clinical outcomes.^22,34 Innovations such as DTx and smart home platforms present efficient and accessible solutions, offering a viable alternative to conventional care models. Other AI-powered platforms, such as ChatGPT, enhance patient education and engagement, providing reliable information for better decision-making.³¹ In short, AI enables precise monitoring and intervention, leading to better resource allocation and health outcomes.^28,32

Clinical Decision Support

Characteristic Features

Technologies incorporating Artificial Intelligence (AI) are increasingly integrated into clinical decision support systems (CDSSs) to enhance diabetes diagnosis, management, and treatment. From a lifestyle medicine perspective, these AI-powered tools are particularly valuable when they (1) facilitate early diagnosis or risk stratification to prompt timely lifestyle interventions, (2) directly provide or support lifestyle-related recommendations, (3) help visualize patient data in ways that inform lifestyle adjustments, or (4) integrate lifestyle factors into comprehensive management strategies (Table 5). These AI-powered tools analyze diverse data sources, including EHRs, imaging data, and patient histories, to provide actionable insights.

Table 5.

Studies Published on Clinical Decision Support.^a

Reference (Year, Country)	Aim	Design & Population	AI Tools and Methods	Results	Conclusions
Liaw WR et al⁵⁷ (2023, USA)	To evaluate the adoption of an AI-based clinical decision support tool	Mixed methods study with 22 clinicians and staff members	Surveys and interviews assessing ease of use and factors affecting adoption	77% expressed intent to use, with a focus on health improvement and tool accuracy	The tool was generally perceived as easy to use and potentially valuable for enhancing population health and individualized care
Shen J et al³⁵ (2020, China)	To implement AI algorithms in GDM diagnosis with minimal resources	Cohort study with 12,304 pregnant women	Tested nine AI algorithms, including SVM and RF, with 5-fold cross-validation and AUROC evaluation	SVM showed the highest accuracy and specific results; effective low-cost diagnostic tool	SVM can effectively diagnose GDM with minimal equipment, making it ideal for resource-limited settings
Xiang Y et al³⁶ (2023, China)	To introduce a noninvasive diabetes diagnostic method using portable equipment	Retrospective cohort with 165 subjects	RF analyzing fundus photography and TCM features like tongue and pulse	RF achieved 0.85 accuracy; effective and noninvasive approach for diabetes diagnosis	The method shows promise for diagnosis in areas with limited medical resources, inspired by TCM
Liu Y et al³⁷ (2020, China)	To apply AI in diagnosing diabetes through neural networks	Cross-sectional study with 650 groups of patients	ANN and backpropagation neural network using MATLAB for diabetes diagnosis and feature extraction	Achieved a 92% correct diagnosis rate	AI applications show utility in aiding diabetes diagnosis and improving diagnostic accuracy
Yoon S et al⁴¹ (2024, Singapore)	To explore clinician perspectives on an AI-enabled prescribing tool	Semistructured interviews with 13 clinicians	Focus on AI-generated drug recommendations and complication risk predictions in diabetes care	Clinicians found AI tool supportive for decision-making, especially for complex cases with comorbidities	AI-CDSSs like APA can enhance clinical practice, particularly in managing complex diabetes cases and facilitating better doctor-patient communication
Yashar MM et al³⁸ (2024, Turkey)	To detect T2D via AI analysis of pectoral muscles on DBT	Retrospective cohort with 11,594 DBT images from 287 women	EfficientNetB5 CNN for image analysis and classification	Achieved 92% accuracy with high specificity and sensitivity	AI-integrated imaging can accurately detect diabetes, offering a noninvasive diagnostic tool
Wang L et al⁴³ (2020, China)	To create a knowledge graph for diabetes management support	Data extracted from literatura	“igraph,” “shiny,” and “ggplot2” in R to model and visualize data related to diabetes and its complications	Knowledge graph highlights lifestyle factors linked to diabetes complications	EBM-based knowledge graphs can support clinical decisions and improve diabetes management by providing high-quality, evidence-based information
Pei X et al⁴⁰ (2022, China)	To assess AI screening for DR in T2D patients	Cross-sectional study with 549 T2D patients	Bayesian Belief Network analysis to evaluate screening accuracy and specificity using various AI tools	EyeWisdom®DSS showed strong sensitivity and specificity for DR screening	AI-based DR screening systems are effective and valuable, especially in resource-limited areas
Derozier V et al⁵⁸ (2019, France)	To develop AI methods for displaying glycemic data effectively	Retrospective cohort with data from 62 patients	ML to analyze glycemic patterns; WEKA clustering to visualize data using color scales	Demonstrated effective display of glycemic patterns for clinical consultations	AI-driven visualization tools can significantly enhance the review and understanding of glycemic data during clinical appointments
BAVIERA-MARTINEZ C et al⁵⁹ (2024, Spain)	To automate CGM data management with an interactive dashboard	Data integrated from multiple sources and visualized on a dashboard	K-means clustering for patient glucose profile classification; continuous updates through Power BI and Timescale	Facilitated real-time CGM data analysis and insights for personalized diabetes management	The proposed system offers innovative tools for proactive diabetes management, enabling effective clinical and resource management
Tarumi S et al⁶⁰ (2021, USA)	To enhance chronic disease care using AI	Cohort study with 27,904 diabetes patients	Stacked models using SVR and LR with integration into a CDSS	Improved prediction accuracy, effectively supporting clinical decision-making	AI-driven CDSS tools successfully integrate with EHRs, offering adaptable frameworks for enhancing chronic care
Hu W et al⁴² (2024, Australia)	To evaluate AI-based DR screening cost-effectiveness	RCT involving over 1.26 million Australians	Markov model and TreeAge Pro for analysis; scenarios A, B, C for intervention cost-effectiveness	Universal AI-based screening most effective and cost-saving; significant impact on reducing blindness in diabetic populations	AI-based DR screenings are both cost-effective and beneficial among Indigenous and non-Indigenous Australians, supporting widespread adoption
Oh SH et al³⁹ (2022, South Korea)	To apply Q-learning for medication recommendations in T2D patients	Cohort study with 14,934 hypertension patients with T2D	Q-learning reinforcement learning model for optimizing treatment based on patient EHRs	Effectively predicted medication strategies, lowering BP and improving treatment outcomes	Reinforcement learning aids in personalized and efficient treatment recommendations, reducing HTN complications
Khalilnejad A et al⁶¹ (2024, USA)	To predict uncontrolled diabetes in RDMP participants using ML	Cohort with over 200,000 T2D participants	LightGBM models trained with temporal data; 5-fold cross-validation for model tuning	High precision and recall in identifying at-risk participants; model performance enhanced with increased participant data	Proactive ML models identify at-risk diabetes patients effectively, supporting timely and personalized interventions within remote monitoring programs
Hoyos W et al⁶² (2023, Columbia)	To develop an AI model for early diabetes detection	Cross-sectional study with fuzzy cognitive maps	Fuzzy cognitive maps to simulate risk factor interactions and enhance clinical decision-making	Achieved 95% accuracy in early diabetes detection through variable interaction analysis	Fuzzy cognitive maps provide valuable insights into diabetes risk dynamics and serve as effective support tools in clinical settings
Fu X et al⁶³ (2023, China)	To compare ML methods for blood glucose prediction in T2D	Retrospective cohort with 273 patients	Multiple ML algorithms, including LR, RF, and XGBoost, for predictive accuracy evaluation	XGBoost achieved highest accuracy and predictive values among methods tested	XGBoost offers clinical decision-making support and improves patient lifestyle guidance to manage glycemic events
Seixas AA et al⁶⁴ (2017, USA)	To assess diabetes prevalence and identify low-prevalence lifestyle combinations	Retrospective cohort with over 288,000 NHIS participants	BBN with Tree Augmented Naïve Bayes for lifestyle impact analysis on diabetes prevalence	Healthy lifestyles combining specific activities significantly reduced diabetes prevalence, particularly in Blacks	Personalized lifestyle interventions can substantially reduce diabetes disparities, warranting tailored public health recommendations
Srinivasu PN et al⁶⁵ (2024, Saudi Arabia)	To create an explainable AI system for glucose monitoring and decision-making	Bi-LSTM and CNN for glucose analysis through spectrogram images	Bi-LSTM and CNN for classifying spectrogram images based on glucose levels	Achieved 96.44% classification accuracy; effective in interpreting glucose level variations	Explainable AI enhances understanding and decision-making in glucose monitoring, making precision healthcare more accessible and effective

^aAbbreviations: AI - Artificial Intelligence; AI-CDSS - Artificial Intelligence enabled Clinical Decision Support System; ANN - Artificial Neural Network; AUROC - Area Under the Receiver Operating Characteristic Curve; BBN - Bayesian Belief Network; Bi-LSTM - Bidirectional Long Short-Term Memory; BP - Blood Pressure; CGM - Continuous Glucose Monitoring; CDSS - Clinical Decision Support System; CNN - Convolutional Neural Network; DBT - Digital Breast Tomosynthesis; DR - Diabetic Retinopathy; EBM - Evidence-Based Medicine; EHR - Electronic Health Record; GDM - Gestational Diabetes Mellitus; HTN - Hypertension; LightGBM - Light Gradient Boosting Machine; LR - Logistic Regression; ML - Machine Learning; NB - Naïve Bayes; NHIS - National Health Interview Survey; Power BI - Power Business Intelligence; RCT - Randomized Controlled Trial; RF - Random Forest; RDMP - Remote Diabetes Monitoring Program; SVM - Support Vector Machine; SVR - Support Vector Regression; TCM - Traditional Chinese Medicine; T2D - Type 2 Diabetes; WEKA - Waikato Environment for Knowledge Analysis; XGBoost - Extreme Gradient Boosting.

Evidentiary studies spanned Australia, China, Colombia, France, Saudi Arabia, Singapore, South Korea, Spain, Turkey, and the U.S. from 2017 to 2024. These studies focused on using AI to improve diabetes diagnosis, treatment recommendations, and management strategies. The research designs included RCTs, mixed methods studies, prospective cohort studies, and cross-sectional studies, with sample sizes ranging from 62 to over 1.26 million participants. The AI tools and methods included SVM, RF, ANN, CNN, reinforcement learning, and Bayesian belief networks. Data sources encompassed EHRs, imaging data, patient histories, and literature.

There are six main categories of AI/ML techniques used for clinical decision support. Shen et al³⁵ used SVM learning to diagnose GDM with high accuracy, making it ideal for resource-limited settings. Random forest methods can be applied to analyze fundus photography and traditional Chinese medicine (TCM) features for noninvasive diabetes diagnosis.³⁶ Notably, Liu et al³⁷ utilized ANN for diabetes diagnosis, achieving a 92% correct diagnosis rate, and Yashar et al³⁸ employed EfficientNetB5 CNN for diabetes detection using digital breast tomosynthesis images with incidental analysis of pectoral muscle. Of further interest, Oh et al³⁹ implemented reinforcement learning (Q-learning) for personalized medication recommendations in patients with T2D. Lastly, Bayesian belief networks have been applied for diabetic retinopathy screening and lifestyle impact analysis.⁴⁰

Key Findings

AI-driven CDSSs contribute to lifestyle-integrated diabetes care in several key ways. Firstly, enhancing early detection and diagnosis is a critical precursor to initiating lifestyle interventions. Studies like Shen et al³⁵ (GDM diagnosis using SVM), and Yashar et al³⁸ (T2D detection from DBT images) and Hoyos W et al⁶² (early diabetes detection with fuzzy cognitive maps) demonstrate AI’s capacity for accurate and often noninvasive diagnosis, enabling earlier engagement in lifestyle modification programs. Similarly, AI-based screening for complications, such as diabetic retinopathy (Pei X et al⁴⁰; Fu X et al⁶⁴), can underscore the urgency for improved lifestyle management.

Secondly, some CDSSs directly support or analyze lifestyle factors. For example, Wang L et al.⁴³ developed a knowledge graph highlighting lifestyle factors linked to diabetes complications, directly informing clinical advice. Seixas AA et al⁶⁴ utilized BBN to explicitly analyze the impact of lifestyle combinations on diabetes prevalence, providing evidence for personalized lifestyle recommendations. Fu X et al⁶³ reported that XGBoost models for blood glucose prediction could improve patient lifestyle guidance to manage glycemic events.

Thirdly, AI tools that enhance the visualization and interpretation of patient data can empower both clinicians and patients to make informed lifestyle choices. Derozier V et al.⁵⁸ demonstrated effective glycemic pattern display for clinical consultations, while BAVIERA-MARTINEZ C et al.⁵⁹ developed an interactive dashboard for real-time CGM data analysis, both facilitating a better understanding that can drive lifestyle adjustments. Finally, even CDSSs with broader aims can be relevant to lifestyle medicine. Tools predicting uncontrolled diabetes (Khalilnejad A et al⁶¹) can help clinicians identify patients needing intensified lifestyle support. Clinician-facing tools (Liaw WR et al.⁵⁷; Yoon S et al.⁴¹) can improve decision-making; their value for lifestyle medicine increases if they incorporate lifestyle data or prompt discussions about non-pharmacological options. Even medication optimization tools (Oh SH et al³⁹) could indirectly support lifestyle medicine by stabilizing a patient or freeing up clinician time for lifestyle counseling.

Discussion

The integration of AI into comprehensive diabetes care heralds a paradigm shift with substantial implications. Our synthesis indicates that AI is being applied across the continuum of lifestyle management, from predicting risk to personalize preventative lifestyle advice (e.g., Naveed et al,¹⁴ Kumar et al¹⁸), to delivering tailored guidance on nutrition and physical activity (e.g., Kannenberg et al,²⁵ Bul et al²⁶), enabling sophisticated remote monitoring and self-management support (e.g., Shamanna et al,²⁸ Helal et al³⁰), and informing clinical decisions pertinent to lifestyle interventions (e.g., Wang L et al,⁴³ Seixas AA et al⁶⁴). A key theme emerging is AI’s capacity to process diverse, complex datasets—from CGM and wearables to EHRs—to offer personalized and timely interventions that were previously unachievable at scale. For instance, the ability of AI to identify subtle patterns in glycemic responses to specific foods or activities (as seen in personalized guidance studies) offers a clear advantage over generic lifestyle advice.

AI’s capabilities align well with the novel dysglycemia-based chronic disease (DBCD) model,⁴⁴ which is pathophysiologically based and emphasizes a continuum of preventive care from insulin resistance (stage 1) to prediabetes (stage 2) to T2D (stage 3) to vascular complications (stage 4). The DBCD model stresses the importance of early intervention at all chronic disease stages through structured lifestyle changes, with judicious pharmacotherapy/procedures at later stages. AI can enhance these efforts by providing precise risk assessments and facilitating early and personalized interventions (Figure 1). Other works in the literature also underscore AI’s potential in improving outcomes through its ability to interpret complex data sets,⁴⁵ thus offering insights into the management of T2D and the prevention of micro- and macrovascular complications, particularly cardiovascular events.⁴⁶

Figure 1.

Examples of Artificial Intelligence Tools for Integration in each State of Dysglycemia-Based Chronic Disease Model.* * The DBCD model is comprised of four stages as depicted to emphasize preventive care. Abbreviations: DBCD - dysglycemia-based chronic disease.

While promising, the evidence base reviewed herein has several limitations that temper immediate widespread application. Many studies rely on region-specific or small datasets, limiting the generalizability of their findings (e.g., some studies in Tables 3 and 4 with smaller cohorts), involved region-specific or relatively small datasets, raising questions about their generalizability across diverse populations and healthcare settings with varying technological infrastructures and socioeconomic realities. The “black-box” nature of some AI models, particularly DL algorithms, poses challenges for interpretability, training, and clinical adoption. Healthcare professionals (HCPs) may hesitate to trust AI recommendations without transparent explanations of how decisions are made. The integration of AI tools into existing healthcare systems faces technical, ethical, and regulatory hurdles, including data privacy concerns and the need for robust validation frameworks. Fourth, the complexity of AI models and the requirement for extensive computational resources can limit scalability, particularly in resource-constrained settings. Fifth, ethical concerns regarding data privacy and patient consent when using AI-driven systems need to be addressed. Finally, while AI models often achieve high accuracy in controlled research settings, their long-term efficacy and real-world performance in diverse clinical environments remain understudied.

Future research on AI in diabetes care will need to address these limitations. Larger, multi-site trials with diverse populations are needed to validate AI tools and establish their generalizability and long-term effectiveness in real-world settings for lifestyle medicine. It will be essential to develop standardized protocols for data collection and AI model validation to ensure consistency and accuracy. Innovations in AI technology should focus on improving user interface design to enhance the accessibility and usability of AI tools for both patients and HCPs. Collaborative efforts between technologists, HCPs, and policymakers will be needed to effectively integrate AI solutions into existing healthcare frameworks. Furthermore, emphasis on ethical AI deployment, including transparent data practices and patient-centered approaches, will be critical to fostering trust and promoting widespread adoption.

This narrative review has several limitations. Firstly, the search was primarily confined to the PubMed database, which, while comprehensive for biomedical literature, may have omitted relevant studies published in dedicated AI/computer science conference proceedings or journals not indexed in PubMed. Secondly, as a narrative review, it does not employ a systematic, reproducible protocol for study selection and data, potentially introducing selection bias. The qualitative synthesis, while aiming for a broad overview, is inherently subjective to some extent. Finally, the focus on English-language publications might have excluded relevant research from other regions. These limitations mean that while this review provides a broad overview of AI in lifestyle medicine for diabetes, it is not an exhaustive account of all existing literature or a quantitative summary of effects.

In conclusion, AI holds considerable promise for revolutionizing lifestyle medicine within diabetes care. This review highlights AI’s emerging capacity to enhance risk prediction for proactive lifestyle interventions, deliver highly personalized guidance for diet and physical activity, empower patients through sophisticated remote self-management tools, and inform clinical decisions pertinent to lifestyle modification. To fully realize this potential, future efforts must focus on robust validation in diverse real-world settings, ensuring equitable access and algorithmic fairness, and seamlessly integrating these powerful tools into patient-centered lifestyle management pathways. Addressing these challenges will be key to harnessing AI to meaningfully improve outcomes for the millions affected by diabetes through enhanced lifestyle medicine.

Footnotes

Declaration of Conflicting Interests

The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: JM reports receiving honoraria for lectures from Abbott Nutrition and Merck and serves on the advisory boards of Abbott Nutrition and Twin Health. JPGR and AS do not have conflicts to declare.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Juan P. González-Rivas

References

Zhou

Rayner

Gregg

, et al. Worldwide trends in diabetes prevalence and treatment from 1990 to 2022: a pooled analysis of 1108 population-representative studies with 141 million participants. Lancet. 2024;404(10467):2077-2093.

Care

. Standards of care in diabetes—2023. Diabetes Care. 2023;46:S1-S267.

Nadal

Angkurawaranon

Singh

, et al. Effectiveness of behaviour change techniques in lifestyle interventions for non-communicable diseases: an umbrella review. BMC Public Health. 2024;24(1):3082.

Ames

Karlsen

Sundermeir

, et al. Lifestyle medicine implementation in 8 health systems: protocol for a multiple case study investigation. JMIR Res Protoc. 2024;13(1):e51562.

Kushner

Mechanick

. Lifestyle medicine: a manual for clinical practice. J Family Med Prim Care. 2016;12(2):208-212.

Davenport

Kalakota

. The potential for artificial intelligence in healthcare. Future Healthc J. 2019;6(2):94-98.

Alowais

Alghamdi

Alsuhebany

, et al. Revolutionizing healthcare: the role of artificial intelligence in clinical practice. BMC Med Educ. 2023;23(1):689.

Mackenzie

Sainsbury

CAR

Wake

. Diabetes and artificial intelligence beyond the closed loop: a review of the landscape, promise and challenges. Diabetologia. 2024;67(2):223-235.

Lee

Kim

Jun

, et al. An integrated digital health care platform for diabetes management with AI-based dietary management: 48-week results from a randomized controlled trial. Diabetes Care. 2023;46(5):959-966.

10.

Wang

SCY

Nickel

Venkatesh

Raza

Kvedar

. AI-based diabetes care: risk prediction models and implementation concerns. npj Digit Med. 2024;7(1):36.

11.

Huang

Zheng

. Application of artificial intelligence in diabetes education and management: present status and promising prospect. Front Public Health. 2020;8:173.

12.

Das

Rahman

Talukder

. Determinants of developing cardiovascular disease risk with emphasis on type-2 diabetes and predictive modeling utilizing machine learning algorithms. Medicine. 2024;103(49):e40813.

13.

Hsieh

Sun

Lin

Hsieh

Hsu

Kao

. The performance of different artificial intelligence models in predicting breast cancer among individuals having type 2 diabetes mellitus. Cancers (Basel). 2019;11(11):1751.

14.

Naveed

Kaleem

Keshavjee

Guergachi

. Artificial intelligence with temporal features outperforms machine learning in predicting diabetes. PLOS Digit Health. 2023;2(10):e0000354.

15.

Choi

Park

, et al. Comparisons of the prediction models for undiagnosed diabetes between machine learning versus traditional statistical methods. Sci Rep. 2023;13(1):13101.

16.

Tsai

Lee

, et al. Prediction of the risk of developing end-stage renal diseases in newly diagnosed type 2 diabetes mellitus using artificial intelligence algorithms. BioData Min. 2023;16(1):8.

17.

Maqbool

Bajwa

Maqbool

Ramzan

Chishty

. A smart sensing technologies-based intelligent healthcare system for diabetes patients. Sensors. 2023;23(23):9558.

18.

Kumar

Ang

Png

, et al. Automated machine learning (AutoML)-derived preconception predictive risk model to guide early intervention for gestational diabetes mellitus. Int J Environ Res Publ Health. 2022;19(11):6792.

19.

Kim

Yoon

Lee

, et al. Artificial intelligence‐based body composition analysis using computed tomography images predicts both prevalence and incidence of diabetes mellitus. J Diabetes Investig. 2024;16(2):272–284.

20.

González-Martín

Torres-Mata

Cazorla-Rivero

, et al. An artificial intelligence prediction model of insulin sensitivity, insulin resistance, and diabetes using genes obtained through differential expression. Genes. 2023;14(12):2119.

21.

Wändell

Carlsson

Wierzbicka

, et al. A machine learning tool for identifying patients with newly diagnosed diabetes in primary care. Prim Care Diabetes. 2024;18(5):501-505.

22.

Ansari

Harris

Hosseinzadeh

Zwar

. Application of artificial intelligence in assessing the self-management practices of patients with type 2 diabetes. Healthcare. 2023;11:903.

23.

Kim

Lee

Choi

, et al. Identification of healthy and unhealthy lifestyles by a wearable activity tracker in type 2 diabetes: a machine learning-based analysis. Endocrinol Metab. 2022;37(3):547-551.

24.

Baum

Scarpa

Bruzelius

Tamler

Basu

Faghmous

. Targeting weight loss interventions to reduce cardiovascular complications of type 2 diabetes: a machine learning-based post-hoc analysis of heterogeneous treatment effects in the Look AHEAD trial. Lancet Diabetes Endocrinol. 2017;5(10):808-815.

25.

Kannenberg

Voggel

Thieme

, et al. Unlocking potential: personalized lifestyle therapy for type 2 diabetes through a predictive algorithm-driven digital therapeutic. J Diabetes Sci Technol. 2024;30:19322968241266820.

26.

Bul

Holliday

Bhuiyan

MRA

Clark

CCT

Allen

Wark

. Usability and preliminary efficacy of an artificial intelligence–driven platform supporting dietary management in diabetes: mixed methods study. JMIR Hum Factors. 2023;10:e43959.

27.

Graham

Pitter

Hori

Stein

Branch

. Weight loss in a digital app-based diabetes prevention program powered by artificial intelligence. Digit Health. 2022;8:20552076221130620.

28.

Shamanna

Joshi

Dharmalingam

, et al. Digital twin in managing hypertension among people with type 2 diabetes: 1-year randomized controlled trial. JACC (J Am Coll Cardiol): Advances. 2024;3(9_Part_2):101172.

29.

Joshi

Shamanna

Dharmalingam

, et al. Digital twin-enabled personalized nutrition improves metabolic dysfunction-associated fatty liver disease in type 2 diabetes: results of a 1-year randomized controlled study. Endocr Pract. 2023;29(12):960-970.

30.

Helal

Cook

Schmalz

. Smart home-based health platform for behavioral monitoring and alteration of diabetes patients. J Diabetes Sci Technol. 2009;3(1):141-148.

31.

Hernandez

Gonzalez

AEV

Polianovskaia

, et al. The future of patient education: AI-driven guide for type 2 diabetes. Cureus. 2023;15(11):e48919.

32.

Williams

Bashan

Kruse

Sritharan

Hodish

. Time in range in patients with type 2 diabetes who are long‐term users of d‐Nav®, an artificial intelligence‐driven technology for autonomous titration of insulin dosing. Diabetes Obes Metabol. 2023;25(12):3845-3848.

33.

Nimri

Battelino

Laffel

, et al. Insulin dose optimization using an automated artificial intelligence-based decision support system in youths with type 1 diabetes. Nat Med. 2020;26(9):1380-1384.

34.

Nayak

Vakili

Nayak

, et al. Use of voice-based conversational artificial intelligence for basal insulin prescription management among patients with type 2 diabetes: a randomized clinical trial. JAMA Netw Open. 2023;6(12):e2340232.

35.

Shen

Chen

Zheng

, et al. An innovative artificial intelligence–based app for the diagnosis of gestational diabetes mellitus (GDM-AI): development study. J Med Internet Res. 2020;22(9):e21573.

36.

Xiang

Shujin

Hongfang

, et al. Artificial intelligence‐based diagnosis of diabetes mellitus: combining fundus photography with traditional Chinese medicine diagnostic methodology. BioMed Res Int. 2021;2021(1):5556057.

37.

Liu

. Artificial intelligence–based neural network for the diagnosis of diabetes: model development. JMIR Med Inform. 2020;8(5):e18682.

38.

Yashar

Izci

Gungoren

Eren

Mert

Durur-Subasi

. Can artificial intelligence detect type 2 diabetes in women by evaluating the pectoral muscle on tomosynthesis: diagnostic study. Insights Imaging. 2024;15(1):99.

39.

Lee

Park

. Precision medicine for hypertension patients with type 2 diabetes via reinforcement learning. J Personalized Med. 2022;12(1):87.

40.

Pei

Yao

Yang

, et al. Efficacy of artificial intelligence-based screening for diabetic retinopathy in type 2 diabetes mellitus patients. Diabetes Res Clin Pract. 2022;184:109190.

41.

Yoon

Goh

Lee

, et al. Assessing the utility, impact, and adoption challenges of an artificial intelligence–enabled prescription advisory tool for type 2 diabetes management: qualitative study. JMIR Hum Factors. 2024;11:e50939.

42.

Joseph

, et al. Population impact and cost-effectiveness of artificial intelligence-based diabetic retinopathy screening in people living with diabetes in Australia: a cost effectiveness analysis. eClinicalMedicine. 2024;67:102387.

43.

Wang

Xie

Han

, et al. Construction of a knowledge graph for diabetes complications from expert-reviewed clinical evidences. Comput Assist Surg. 2020;25(1):29-35.

44.

Mechanick

Farkouh

Newman

Garvey

. Cardiometabolic-based chronic disease, adiposity and dysglycemia drivers: JACC state-of-the-art review. J Am Coll Cardiol. 2020;75(5):525-538.

45.

Gan

Wang

Liu

. Proteomics-based mortality prediction modeling in type 2 diabetes: new promise for personalized treatment and prevention. BMC Med. 2025;23(1):1-3.

46.

Min

, et al. Effect of artificial intelligence-based health education accurately linking system (AI-HEALS) for type 2 diabetes self-management: protocol for a mixed-methods study. BMC Public Health. 2023;23(1):1325.

47.

Okada

Hashimoto

Goto

, et al. A machine learning–based predictive model to identify patients who failed to attend a follow-up visit for diabetes care after recommendations from a national screening program. Diabetes Care. 2022;45(6):1346-1354.

48.

Borzouei

Soltanian

. Application of an artificial neural network model for diagnosing type 2 diabetes mellitus and determining the relative importance of risk factors. Epidemiol Health. 2018;40:e2018007.

49.

Elhadd

Mall

Bashir

, et al.. Artificial Intelligence (AI) based machine learning models predict glucose variability and hypoglycaemia risk in patients with type 2 diabetes on a multiple drug regimen who fast during ramadan (The PROFAST–IT Ramadan study). Diabetes Res Clin Pract. 2020;169:108388.

50.

Neri-Rosario

Martínez-López

Esquivel-Hernández

, et al. Dysbiosis signatures of gut microbiota and the progression of type 2 diabetes: a machine learning approach in a Mexican cohort. Front Endocrinol. 2023;14:1170459.

51.

Kim

Park

Lee

. Explainable artificial intelligence on life satisfaction, diabetes mellitus and its comorbid condition. Sci Rep. 2023;13(1):11651.

52.

Gallardo-Rincón

Ríos-Blancas

Ortega-Montiel

, et al. MIDO GDM: an innovative artificial intelligence-based prediction model for the development of gestational diabetes in Mexican women. Sci Rep. 2023;13(1):6992.

53.

Watanabe

Eguchi

Sakurai

Yamamoto

Mori

. Prediction of gestational diabetes mellitus using machine learning from birth cohort data of the Japan Environment and Children’s Study. Sci Rep. 2023;13(1):17419.

54.

Yao

Chae

MacLean

, et al. SynthA1c: towards clinically interpretable patient representations for diabetes risk stratification. In: International Workshop on PRedictive Intelligence in MEdicine. Berlin, Germany: Springer; 2023:46-57.

55.

Haneef

Kab

Hrzic

, et al. Use of artificial intelligence for public health surveillance: a case study to develop a machine Learning-algorithm to estimate the incidence of diabetes mellitus in France. Arch Public Health. 2021;79:1-13.

56.

Hennebelle

Ismail

Materwala

Al Kaabi

Ranjan

Janardhanan

. Secure and privacy-preserving automated machine learning operations into end-to-end integrated IoT-edge-artificial intelligence-blockchain monitoring system for diabetes mellitus prediction. Comput Struct Biotechnol J. 2024;23:212-233.

57.

Liaw

Ramos Silva

Soltero

Krist

Stotts

. An assessment of how clinicians and staff members use a diabetes artificial Intelligence prediction tool: mixed methods study. JMIR AI. 2023;2:e45032.

58.

Derozier

Arnavielhe

Renard

Dray

Martin

. How knowledge emerges from artificial intelligence algorithm and data visualization for diabetes management. J Diabetes Sci Technol. 2019;13(4):698-707.

59.

Baviera-Martineza

Martinez-Millana

de Borja Lopez-Casanova

. Integrating automation, interactive visualization, and unsupervised learning for enhanced diabetes management. Stud Health Technol Inf. 2024;316:1699-1703.

60.

Tarumi

Takeuchi

Chalkidis

, et al. Leveraging artificial intelligence to improve chronic disease care: methods and application to pharmacotherapy decision support for type-2 diabetes mellitus. Methods Inf Med. 2021;60(S 01):e32-e43.

61.

Khalilnejad

Sun

Kompala

Painter

James

Wang

. Proactive identification of patients with diabetes at risk of uncontrolled outcomes during a diabetes management program: conceptualization and development study using machine learning. JMIR Form Res. 2024;8(1):e54373.

62.

Hoyos

Ruiz-Pérez

. Artificial intelligence model for early detection of diabetes. Biomedica. 2023;43:110-125.

63.

Wang

Cates

, et al. Implementation of five machine learning methods to predict the 52-week blood glucose level in patients with type 2 diabetes. Front Endocrinol. 2023;13:1061507.

64.

Seixas

Henclewood

Langford

McFarlane

Zizi

Jean-Louis

. Differential and combined effects of physical activity profiles and prohealth behaviors on diabetes prevalence among blacks and whites in the US population: a novel Bayesian belief network machine learning analysis. J Diabetes Res. 2017;2017(1):5906034.

65.

Srinivasu

Ahmed

Hassaballah

Almusallam

. An explainable Artificial Intelligence software system for predicting diabetes. Heliyon. 2024;10(16):e36112.

Artificial Intelligence Enabled Lifestyle Medicine in Diabetes Care: A Narrative Review

Abstract

Keywords

Introduction

Methods

Results

Predictive AI Models

Characteristic Features

Key Findings

Personalized Lifestyle Guidance

Characteristic Features

Key Findings

Remote Monitoring and Self-Management

Characteristic Features

Key Findings

Clinical Decision Support

Characteristic Features

Key Findings

Discussion

Footnotes

Declaration of Conflicting Interests

Funding

ORCID iD

References